Fast Approximate Matching of Cell-Phone Videos for Robust Background Subtraction

摘要

We identify a novel instance of the background subtraction problem thatfocuses on extracting near-field foreground objects captured using handheldcameras. Given two user-generated videos of a scene, one with and the otherwithout the foreground object(s), our goal is to efficiently generate an outputvideo with only the foreground object(s) present in it. We cast this challengeas a spatio-temporal frame matching problem, and propose an efficient solutionfor it that exploits the temporal smoothness of the video sequences. We presenttheoretical analyses for the error bounds of our approach, and validate ourfindings using a detailed set of simulation experiments. Finally, we presentthe results of our approach tested on multiple real videos captured usinghandheld cameras, and compare them to several alternate foreground extractionapproaches.